Regular restore tests
- From: Jeremy Schneider <schneider@xxxxxxxxxxxxxx>
- To: John Hallas <john.hallas@xxxxxxxxxxxxxxxxxx>
- Date: Thu, 13 Apr 2017 19:18:42 +0300
On Thu, 13 Apr 2017 09:46:37 +0100 John Hallas wrote:
As an aside we have a server (well 2 actually - one HPUX and one
Linux) that we use for backup proving.
We have a couple of Oracle homes on each and we recover the backups
from our Commvault repository onto these servers. If we have an
11.2.0.3 database we will recover into whichever OH is installed - it
might be 11.2.0.2 or 11.2.0.4. and we do not run any
upgrade/downgrade scripts All we are testing is that the database can
be restored and that the backup is fully functional
Nice, John!
In my most recent position before this current freelancing spell, I
helped architect our restore testing process. Every week we randomly
picked one of our several hundred production web applications and did a
full out-of-place restore of both the database and application tiers. We
then logged into the restored application and pulled screenshots
showing dates of recent entries (but no sensitive data), proving the
date of the restore. We stored the screenshots as part of the evidence
on our ISO & SAS certifications. I found that auditors really liked
these screenshots too - proving the restore visually without lots of
explanation, unlike log or terminal captures.
One of the counterintuitive principles in devops is that when something
is painful you should consider doing it more instead of less. Because
we did these restores weekly, we got very skilled and fast and reliable
in running them. The restores were also shared across the DBA team so
that we all acquired the skills. Eventually, every step was either
automated (by code in svn) or cut-and-pasted from a change-controlled
wiki.
I was really proud of what we did on that operations team!
-Jeremy
--
http://about.me/jeremy_schneider
##
Ok, I guess I just got so impressed with the size of a 64-bit value that
I was overwhelmed. Consider, for example:
u64 i;
for (i = 1; i != 0; i++);
Now in theory this will count each possible number, but in practice the
machine will die long before it ever finishes.
- George Anzinger on linux-kernel
--
//www.freelists.org/webpage/oracle-l
Other related posts:
